Long-term effects on functional brain networks in adolescents treated for lumbar disc herniation

TL;DR

This study found that adolescents who had surgery for a spinal disc issue showed lasting changes in brain connectivity, even years later.

Contribution

The study reveals long-term brain network changes in adolescents after surgical treatment for lumbar disc herniation.

Findings

Surgically treated adolescents showed distinct brain connectivity 12 years post-treatment.

These changes were not linked to pain or spinal structure differences.

Non-operative treatment and controls did not show similar brain connectivity patterns.

Abstract

Long-term effects of lumbar disc herniation treatment on brain function are poorly understood, and it is unclear when surgery should be recommended over non-operative treatment. The overall aim of the present study was to determine potential long-term effects on brain networks among individuals who received either surgical or non-operative treatment for lumbar disc herniation in adolescence. Brain network connectivity was assessed for individuals who received surgical treatment or non-operative treatment, and controls with no history of lumbar disc herniation. Prior to analysis, brain connectivity measures between groups were determined as main outcome, using functional magnetic resonance imaging. On average 12 years after treatment onset, the surgically treated cohort exhibited distinctly different functional brain connectivity, compared with both non-operative treatment and controls. The difference was neither attributed to self-reported pain, nor lumbar spine morphology. The findings suggest that surgical treatment for lumbar disc herniation in adolescence may be associated with a long-term imprint on the functional brain connectome.

Figures

Self-reported outcomes. Error bars represent +/- 2 SE. (Sign. Code: <0.05 *, <0.01, <0.001 *)

Significant brain network clusters. Permutation testing with F-statistics using the Network-based statistics generated three significant clusters where functional brain connectivity differed significantly (A-C). The size of the highlighted nodes from the cluster is synonymous with the number of connections it has. Frequency of participation for each network was obtained by counting the connections that reached their respective nodes.

Seed-based connectivity analysis. (A) One node per hemisphere (LH/RH) that covered the back area of the primary somatosensory cortex was selected as seed region, illustrated in blue. (B) The connectivity with the remaining 122 nodes were pairwise compared. (C-D) Illustration of significantly different connections to seed between surgery/non-operative (C), as well as surgery/control (D) (p< 0.05, FDR). No significantly different seed connection could be identified by comparing non-operative treatment with controls, nor in the RH seed for the surgery/control comparison. Thus, no illustrations were generated.

Full text

Introduction

As of 2020, long-term low back pain (LBP) remains the leading cause of years lived with disability globally.^ 1 ^ Neuroimaging studies show that long-term pain is associated with structural- and functional changes in the brain.^2,3^ Similarly, normalization of brain aberrations has been identified when long-term pain has been successfully treated, for example, in LBP,^ 4 ^ arthritis,^ 5 ^ and fibromyalgia,^ 6 ^ suggesting that brain abnormalities are not the cause, but a consequence of brain plasticity during persistent pain.

One cause for long-term LBP is lumbar disc herniation (LDH).^ 7 ^ LDH is common in adulthood, but rare in adolescence.^ 8 ^ LDH may be treated surgically or non-surgically.^9,10^ Non-surgical treatment consists mainly of medication, physiotherapy and lifestyle modifications, which has an equally large improvement rate in adults, although remission is often slower compared to surgery.^11,12^

Long-term outcomes of LDH treatments have previously been described in adults,^13?–15^ but knowledge is still lacking when treatments are initiated during adolescence. Lagerbäck et al. previously performed a case/control study on the long-term effects of LDH surgery in adolescence, examining spinal Magnetic Resonance Imaging (MRI) characteristics as well as quality-of-life factors 14 years after treatment.^ 16 ^ Findings suggest that cases had more degenerative signs at the two lower lumbar levels (L4-L5 and L5-S1), more back pain and a lower quality of life compared to controls. In addition, Pontén et al. compared the same surgically treated cohort with individuals who received non-operative treatment and found similar degrees of lumbar degeneration and patient-reported outcomes at the time of diagnosis, and at long-term follow-up.^ 17 ^

Functional MRI (fMRI) of the brain, performed during a task free condition (so called resting-state fMRI), can be used to determine connectivity of intrinsic brain networks, referred to as the functional connectome, and allows for detection of network changes linked to disease. Long-term back pain,^ 18 ^ and spinal cord injuries,^ 19 ^ have previously been associated with altered connectivity in resting-state functional brain networks, but knowledge is lacking for long-term follow-ups. Over the last decade, researchers have utilized resting-state fMRI to investigate brain-related changes when suffering from LDH,^ 20 ^ and subsequent non-operative treatments.^21,22^ Using resting state fMRI and graph-based analyses, these studies suggest aberrant whole-brain network properties in LDH compared to controls.^ 20 ^ Also, preliminary evidence from an fMRI study suggests a possible change in the amplitude of spontaneous low-frequency fluctuations in the brain after treatment with manual therapy.^ 22 ^ To our knowledge, no brain fMRI studies have yet compared different treatment options for LDH.

Here, we used resting-state fMRI to determine the potential long-term effects on the functional connectome among individuals who received either surgical or non-operative treatment for LDH in adolescence, along with controls.

Three hypotheses were prepared to answer the research question.

Hypothesis I: There will be a significant difference in resting-state functional brain connectivity between the three cohorts: surgical, non-operative and control.Hypothesis II: The brain region representing the lower back area in the primary somatosensory cortex will display cohort-specific resting-state functional connectivity profiles to the rest of the brain.Hypothesis III: Severe intervertebral disc morphology, measured with the Pfirrmann grading score will correlate with functional connectivity measures in resting-state brain networks.

Materials and methods

Ethical approval of this study was received from the Ethical Review Board in Stockholm (dnr: 2018/29931/1 and 2019-01713). The analyses were pre-registered and are available at OSF via https://osf.io/36qxc.The dataset had not been opened during the preregistration. Upon opening of the dataset, some changes in hypotheses were made before proceeding into analysis: Changes in hypotheses from preregistration included analyzing non-operative and surgically treated participants separated, and not pooled. Additionally, at the time of analysis there were no clinical cutoffs for a patient’s status of the lumbar back. Therefore, we used a pre-defined threshold of Pfirrmann score IV and above in the L4-L5 and L5-S1 lumbar back area to dichotomize participants to having “severe intervertebral disc degeneration” or not.^ 23 ^

Participants

All participants gave oral and written informed consent. Exclusion criteria for all participants were: contraindications for undergoing MRI; serious health problems and/or any kind of substance abuse. An additional exclusion criterion for the non-operative cohort was a history of lumbar surgery.

Individuals treated surgically for LDH during adolescence were identified from the Swedish Spine register (SweSpine). Spinal MRI characteristics of participating individuals (n = 23) were previously compared with age- and gender matched controls (n = 23) to assess the long-term effects of surgery for LDH.^ 16 ^ Structural and resting-state functional brain MRI data at the time of follow-up (May 2019–Jan 2020) were available from all participants in both cohorts.

Individuals with non-operative treatment for LDH, diagnosed during adolescence, were identified based on hospital records at the Karolinska University Hospital Solna, Sweden. A total of 22 individuals were recruited and answered a follow-up survey. Two individuals were excluded because they had undergone surgery. Of the remaining 20 individuals, 14 underwent spinal MRI to compare their degree of degeneration to the degeneration at the time of diagnosis.^ 17 ^ Structural and functional brain MRI data were collected for 12 participants at the time of follow-up (Jun 2021 - Sep 2021) and were thus collected after the first two cohorts (May 2019 – Jan 2020) with no possibility to do a priori stratification on age and sex.

Clinical characteristics

Clinically reported outcomes included disc degeneration assessment using the Pfirrmann classification on sagittal T2-weighted lumbar MRI images.^ 24 ^

Participants self-assessed the Oswestry Disability Index (ODI) survey, EQ VAS level, and Visual Analogue Scale (VAS) estimates of both back and leg pain.

ODI (score 0-100%) is a back-related disability score that ranges between 0% (no disability) to 100% (completely disabled).^ 25 ^

The EQ VAS (score 0-100) is a self-estimation of a person’s state of health. The score “100” represents the best possible condition and “0” the worst possible condition.^ 26 ^

The VAS (0-100 mm) was collected for last week’s back pain and leg pain. The anchor words “No pain” and “Maximum pain” was used for “0” and “100,” respectively.

Image acquisition and preprocessing

Brain MRI data were collected with an 8-channel head coil and a 3.0T scanner (Discovery MR750, GE Healthcare) at MR-center, located in Karolinska University Hospital, Solna, Sweden. For acquisition of functional brain images, gradient echo-planar-imaging (EPI) was used with the following specifications: TR = 2205 ms, TE = 30 ms, flip angle = 80°, FoV = 230 × 230 mm, slices = 42, slice thickness = 3.0 mm^3^ (0.5 mm spacing, axial orientation, phase-encoding direction A/P), number of volumes = 403.

Collected brain images were evaluated in MRIQC,^ 27 ^ and then further preprocessed using fMRIPrep.^ 28 ^ A full description of the image preprocessing pipeline and steps from fMRIPrep are available in the supplementary materials. Motion correction and denoising of preprocessed functional brain imaging data were performed using the 36-nuisance regressor strategy.^29,30^ Subsequently, all images were band-pass filtered (0.008–0.1 Hz).

Computation of functional brain connectivity

Preprocessed functional images were parcellated with FSL into regions of interest (nodes) using the Schaefer 100 atlas.^31,32^ The nodes were further grouped into seven brain networks defined in Yeo et al.^ 33 ^ Additionally, the Harvard-Oxford subcortical parcellation (13 nodes) and King 2019 cerebellum parcellation (10 nodes) were added.^34???–38^ In total, nine brain networks were defined in the analysis. At a nodal level, all pairs were correlated with each other, producing a 123 × 123 nodal connectivity matrix for each individual. To address hypothesis II, we selected the two nodes covering the hip/trunk area of the somatosensory cortex (15 and 66 from the Schaefer/Yeo 100 parcel 7 network scheme) to conduct seed-based correlational analyses to the rest of the brain.

Statistical analyses

Statistical analyses were conducted using MATLAB, Python and SPSS. Python scripts for analyses and illustrations are available at https://github.com/kipain/back2brain-connectivity.

Sex distribution between groups was compared using chi-square test. Age and clinical characteristics were compared between groups using ANOVA. Post hoc analyses were adjusted using two-sided Bonferroni correction with an alpha value of 0.05.

All brain connectivity matrices were fisher z-transformed prior to analysis. For cluster-based statistics addressing hypothesis I and III, a previously described Network-based statistic (NBS) approach was used to identify deviating clusters of connectivity patterns through permutation testing.^ 39 ^ This network-based approach corrects for multiple comparisons with family-wise error rate. The parameters used for the NBS analysis were f-statistics with a preregistered alpha of .05, an f-threshold of 9 (corresponding to a t-threshold of 3), and 10,000 permutations. f_max_ represents the largest f-value found within a cluster. Post-hoc analyses used identical parameters.

The seed-based analyses addressing hypothesis II used FDR-corrected two-sided paired t-tests, comparing the mean differences in connectivity between seed nodes (left and right hemisphere, respectively) and all other regions of the brain.

Results

Clinical Outcomes

Demographics and clinical characteristics are summarized in Table 1. The differences observed in both age and sex for the non-operative cohort were taken into consideration by adding them as covariates in the brain network analyses. Chi-square test found no significant differences in the sex distribution (x^2^ = 4.521, df = 2, p = 0.1). There was however a significant difference in age between the three groups (F(2,63) = 22.4, p < 0.001).

Table 1.: Demographics and clinical characteristics.

Pain during the last week was compared between groups. A significant difference was found for pain in the back (F(2,62) = 7.6, p = 0.001), but not in the leg, (F(2, 62) = 2.7, p = 0.077). In post-hoc pairwise comparisons of back pain, we found that the control group reported significantly lower levels than both treatment cohorts, respectively (Surgery: t(43) = 3.4, p < 0.001, Non-operative: t(40) = 4.1, p < 0.001). However, there was no significant difference in back pain between the treatment cohorts (t(41) = -0.8, p = 0.425).

There was a significant difference between groups for the ODI score, (F(2,63) = 12.9, p < 0.001). Post-hoc pairwise comparisons presented a significantly lower disability score in the control group in comparison with both respective cohorts (Surgery: (t(44) = 5.1, p < 0.001, Non-operative: t(41) = 4.9, p < 0.001). For this comparison, no significant difference could be identified between the treatment cohorts regarding ODI score either (t(41) = −0.5, p = 0.621).

The EQ VAS for self-estimated health was significantly different between groups (F(2,62) = 11.1, p < 0.001). Pairwise comparisons revealed significantly higher health estimations in the control group when compared with surgical treatment (t(43) = -3.1, p = 0.003), as well as non-operative treatment (t(40) = −5.3, p < 0.001). Here, EQ VAS differed significantly between treatment cohorts with the surgical group reported higher health ratings (t(41) = 2.0, p = 0.048).

All significant comparisons survived Bonferroni correction at the alpha level of 0.05, two-sided, except for the significant difference in EQ VAS between surgical and non-operative treatment (Figure 1).

*Self-reported outcomes. Error bars represent +/- 2 SE. (Sign. Code: <0.05 *, <0.01, ** <0.001 **)

Whole-brain connectivity

One significant cluster of brain connectivity was detected, using permutation f-tests from NBS on all groups, after controlling for sex and age (Figure 2(a), n = 7 edges, f_max_ = 11.44, p = 0.020). Post-hoc permutation f-tests for the three combinations of group pairs further found significant connectivity clusters when the surgical treatment cohort was compared with non-operative treatment (Figure 2(b), n = 88 edges, f_max_ = 26.33, p = 0.018) and controls (Figure 2(c), n = 59 edges, f_max_ = 20.74, p = 0.048). However, comparing non-operative treatment with controls resulted in no significant brain connectivity cluster (p = 0.35). The networks with the most prominent participations within these clusters were the somatomotor, frontoparietal, and default mode networks.

Significant brain network clusters. Permutation testing with F-statistics using the Network-based statistics generated three significant clusters where functional brain connectivity differed significantly (A-C). The size of the highlighted nodes from the cluster is synonymous with the number of connections it has. Frequency of participation for each network was obtained by counting the connections that reached their respective nodes.

Seed-based connectivity

The degree of connectivity between our a priori seed region of interest (S1 back area, Figure 3(a)) and the rest of the brain (Figure 3(b)) was different between the surgical group and the non-operative group at 14 significant connections (edges), divided on both right (RH) and left (LH) hemisphere (Figure 3(c)). All edges were significant after adjusting for 122 comparisons (nodes in rest of the brain) per analysis, Additionally, 10 edges to the LH seed were significantly different between the surgical group and controls (Figure 3(d)), but none were significant to the RH seed. No significant difference was observed between the non-operative group and controls after correcting for multiple comparisons.

Seed-based connectivity analysis. (A) One node per hemisphere (LH/RH) that covered the back area of the primary somatosensory cortex was selected as seed region, illustrated in blue. (B) The connectivity with the remaining 122 nodes were pairwise compared. (C-D) Illustration of significantly different connections to seed between surgery/non-operative (C), as well as surgery/control (D) (p< 0.05, FDR). No significantly different seed connection could be identified by comparing non-operative treatment with controls, nor in the RH seed for the surgery/control comparison. Thus, no illustrations were generated.

Brain connectivity and back morphology

A dichotomization based on severity of spinal morphology divided the participants into two equally large groups, pooled across all three groups. However, permutation testing saw no significant differences in brain connectivity.

Discussion

In response to our first hypothesis, we found that a significant cluster of intrinsic brain connectivity separated the three groups (Figure 2). Pairwise statistical comparisons indicate that the main differences stem from the surgically treated cohort as no significant pairwise results could be identified for the non-operative versus control group. NBS clusters, derived from rejecting the null hypothesis that no differences exist between the groups compared, should be interpreted with caution as their properties may depend on other factors, such as signal-to-noise ratios and the chosen threshold parameter. Therefore, we will describe only the participating networks with a large proportion of connections across comparisons. These were the somatomotor, frontoparietal, and default mode network.

The somatomotor network represents primary somatosensory signaling and is therefore central for understanding pain processing, not necessarily in its own right but in connection to pain-related cognition and emotion circuitry.^ 40 ^ The frontoparietal network, including the prefrontal, parietal, and medial prefrontal cortices, is implicated in decision-making by integrating external information with internal representations.^ 41 ^ As such, the frontoparietal network plays a key role in pain modulation, with studies showing that anticipatory activity in this network predicts placebo responses.^ 42 ^ Notably, activation during pain anticipation overlaps with the resting-state frontoparietal network,^ 43 ^ suggesting its significant role in the internal representation of pain. In the present study, the differences in frontoparietal network connectivity between groups may reflect differences in long-term exposure to internal representation of pain, and related differences in neural imprint. In addition, a large number of studies demonstrate that pain affects the connectivity within the default mode network,^44??–47^ and the degree of disruption has been linked to pain severity.^ 48 ^ In our data, the difference in brain connectivity between the surgical and non-operative group was likely not attributed to reported pain levels as the pain at follow-up was reported higher within the non-operative group (Figure 1).

For our second hypothesis, we determined the connectivity between the back area of the primary somatosensory cortex and the rest of the brain. These results furthered our previous findings of group differences reflected in connectivity between the somatomotor and default mode networks (Figure 3). Moreover, the dorsal- and ventral attention networks accounted for some significantly different connections to the somatosensory back seed as well. Both these functional networks are associated with salience signaling (importance attribution),^ 49 ^ but could not be linked to levels of perceived pain at follow-up, and are likely reflecting other aspects of group differences than current pain levels.

Lastly, we hypothesized that differences in spinal morphology would distinguish different functional networks in the brain. However, permutation testing of a group dichotomization, based on lumbar disc severity, did not yield any significant clusters of change within the brain. It has for long been suggested that MRI characteristics of the spine are not associated with pain levels and other behavioral outcomes,^50,51^ and our study further supports the proposal with a novel perspective by addressing functional brain connectivity.

A previous brain MRI study on chronic LBP used gray matter density to distinguish LBP patients from healthy controls with a 76% accuracy.^ 52 ^ Zhou et al. found similar networks to be activated for LDH patients in a resting-state brain fMRI study.^ 53 ^ Furthermore, they propose that changes in the visual network are associated with pain improvements in LDH when treated with Spinal Manipulative Therapy, a non-operative method.^ 53 ^ Du et al. have also examined the effects of Spinal Manipulative Therapy on LDH and suggested alterations in brain function, due to LDH, to be reversible using this treatment method.^ 21 ^ To our knowledge, no comparable studies exist for surgically treating LDH, which makes it difficult to address whether our findings support the reversibility when treated non-operatively or not. As our data did not contain baseline resting-state fMRI data, we were unable to address the involvement of the visual network in LDH improvement. However, the potential alterations in brain function does not appear to have been successfully reversed in the surgically treated group.

Overall, the non-operative cohort exhibited only a few significant differences in functional connectivity compared to controls. Contrastingly, the largest differences in brain connectivity were found between surgical and non-operative treatment. With the lack of baseline brain fMRI data, we are unable to further determine any if these differences and similarities hold any therapeutic value. If one would speculate, the surgically treated group may display these differences due to different baseline severities. It is common practice to recommend surgery for patients suffering from large disc herniations as they produce worse radiculopathies. If so, this would suggest that a long-term imprint in the functional connectome may have occurred due to a more severe baseline. Alternatively, it may occur when exposed to a higher degree of lumbar spine complications.

To a certain extent, our findings of group differences at 12-year follow-up contradict previous findings by Du, and Seminovicz et al., where altered functional brain connectivity following chronic low back pain was reverted upon successful treatment. However, unlike that population, the present study examines treatments during adolescence, in which a higher degree of cerebral plasticity may cause more pronounced functional reorganizations in response to back pain. This uncertainty could be resolved by replicating the procedure presented here on a population treated for LDH in adulthood and comparing the findings. Another aspect worth highlighting is that the non-operative treatment cohort has their spinal tissue intact. If the present differences between groups are the result from a long-term adaptation to an invasive procedure, it may be important to monitor such individuals further, to prevent adverse effects from developing. To shed light on the causality, future studies could characterize the functional brain connectome at baseline and compare the long-term outcome between a surgical and non-operative treatment.

Concerning the limitations of the study, we acknowledge the small sample size and urge for verifying replications in larger samples. Furthermore, without baseline functional brain images nor a comparable adult population, we are unable to address the causality of the changes presented here. Future longitudinally designed studies may elucidate whether the differences observed here are in fact imprints from plasticity in a surgically treated group. Lastly, we observed a heterogeneity within the controls as they exhibited varying degenerative signs of their lumbar discs and perceived pain.

To conclude, over a decade after treatment for LDH, we have identified a prominent change in brain communication patterns for those receiving surgery in their adolescence. Future investigations may elucidate whether these differences can be attributed to plasticity or perhaps an adverse effect from the intervention itself.

Supplemental Material

sj-docx-1-mpx-10.1177_17448069251376189 – Supplemental material for Long-term effects on functional brain networks in adolescents treated for lumbar disc herniationSupplemental material, sj-docx-1-mpx-10.1177_17448069251376189 for Long-term effects on functional brain networks in adolescents treated for lumbar disc herniation by Sebastian Blomé, Granit Kastrati, Sebastian Pontén, Martin Jonsjö, Tobias Lagerbäck, Mikael Skorpil, Hans Möller, Maria Lalouni, Peter Fransson, Paul Gerdhem, William Hedley Thompson and Karin Jensen in Molecular Pain